We @citizenlab w/TAG's @maddiestone caught #predator spyware attacks against a prominent pro-democracy Egyptian politician after he announced presidential ambitions.
4/ This kind of exploit delivery through injection DOES NOT require a target to click as our collaborator, the brilliant @maddiestone, points out in her post.
@maddiestone @billmarczak 7/ Ahmed ElTantawy wasn't just targeted with network injection!
He was also targeted with #Predator spyware links in decoy messages sent as texts & over @WhatsApp.
One of the attacks masqueraded as communications from the International Federation for Human Rights @fidh_en
@maddiestone @billmarczak @WhatsApp @fidh_en 8/ This summer the 🇺🇸US hit developer & distributor of #Predator spyware (Cytrox & Intellexa) with blacklisting.
This latest abuse revelation affirms the determination that the spyware continues to fuel human rights abuses.
Here are some more damming revelations as Intellexa, the shady, sanctioned spyware supplier gets exposed by @AmnestyTech & partners.. /1
2/ First, a mercenary spyware myth has just been busted.
Because the leak shows an Intellexa employee directly accessing a customer deployment.
Prior to the #PredatorFiles leak, spyware companies basically always claimed they couldn't access customer deployments & didn't know what was going on there.
They used this to avoid responsibility & claim ignorance when faced with abuses.
3/ And it gets crazier. The leak shows Intellexa casually accessing a core backbone of Predator deployment of a government customer.
Seemingly without the gov's knowledge.
Suggests that Intellexa can look over their shoulder & watch their sensitive targeting.
NEW: 🇨🇳Chinese hackers ran massive campaign by tricking Claude's agentic AI.
Vibe hacking ran 80-90% of the operation without humans.
Massive scale (1000s of reqs/sec).
Agents ran complex multi-step tasks, shepherded by a human.
Long predicted. Welcome to the new world.
Fascinating report by @AnthropicAI 1/
2/ The old cybersecurity pitch: unpatched systems are the threat.
The next generation concern might be unpatched cognition.
The attacker jailbroke the cognitive layer of @anthropic's Claude code, successfully convincing the system of false intent (that it was a security exercise)
3/ One of the key points in @AnthropicAI's report is just how limited the human time required was to run such a large automated campaign.
Obviously powerful stuff, highlighting the impact of orchestration.
And concerning for the #cybersecurity world for all sorts of reasons, ranging from attack scale, adaptability & cost reductions...